AI SAFETY: COMPANIES VS. COUNTRIES CLASH

At a Rest of World event in New York, speakers addressed how nations overlooked in the AI power struggle can maintain regulatory control over safety standards. AI companies prefer embedding their own safety evaluators into deployed models. This approach keeps evaluation in-house and streamlines deployment across markets. However, governments argue this model limits their ability to enforce local safety requirements and standards. Countries adopting these systems face a choice: accept corporate-defined safety measures or build independent evaluation infrastructure. The tension reflects broader concerns about AI governance—whether standards should be set by the companies building systems or by elected authorities. Smaller nations particularly worry about losing leverage. Without their own evaluation capabilities, they become dependent on corporate decisions around content moderation, bias assessment, and risk management. The debate signals an emerging split in how AI safety will be governed globally, with competing visions of who controls the guardrails.